Google Books N-gram Corpus used as a Grammar Checker

In this research we explore the possibility of using a large n-gram corpus (Google Books) to derive lexical transition probabilities from the frequency of word n-grams and then use them to check and suggest corrections in a target text without the need for grammar rules. We conduct several experiments in Spanish, although our conclusions also reach other languages since the procedure is corpus-driven. The paper reports on experiments involving different types of grammar errors, which are conducted to test different grammar-checking procedures, namely, spotting possible errors, deciding between different lexical possibilities and filling-in the blanks in a text.